rsi_today

We use RSI to automatically improve a language model's physics: builds the training environment, data, and checker-based reward, then fine tune a Qwen model with reinforcement learning to draw physically correct frame-by-frame scenes